Is Silence Golden? An Empirical Analysis of Firms that Stop Giving Quarterly Earnings Guidance in

the post Regulation-FD period

Shuping Chen Assistant Professor

Email: shupingc@u.washington.edu

Dawn Matsumoto Associate Professor

Email: damatsu@u.washington.edu

Shiva Rajgopal Herbert Whitten Professor

Email: rajgopal@u.washington.edu

University of Washington Box 353200

Seattle, WA 98195

Abstract:

We investigate a sample of 96 firms that publicly renounced quarterly EPS guidance in the post-FD period (10/2000 to 1/2006). We find that stoppers have poor trailing stock return performance and lower institutional ownership. We document an average negative 4.8% three-day return around the announcement to stop guidance and this reaction is associated with poor future performance. After the elimination of guidance, stock prices lead earnings less but there is no change in overall stock return volatility or analyst following. However, analyst forecast dispersion increases and forecast accuracy decreases following firms’ decision to stop guiding, despite increased disclosures made in earnings press releases.

Current Draft Date: October 2006

We thank Bill Beaver, Bob Bowen, Steve Baginski, Dave Burgstahler, Yonca Ertimur, George Foster, Frank Hodge, Amy Hutton, Ron Kasznik, Ben Lansford, Terry Shevlin, Mark Solimon, workshop participants at Stanford University and the University of Washington, the 2006 FARS mid-year conference, the 2006 UBCOW conference, and the 2006 AAA conference for their comments. The authors thank the UW Accounting Development Fund and the CFO Forum for financial support. In addition, Professor Rajgopal received financial support from the Helen Moore Fellowship and the Herbert Whitten Fellowship in accounting and Professor Matsumoto received support from the William R. Gregory Fellowship.

This is a preprint version of the article. The final version may be found at < http://dx.doi.org/10.1016/j.jacceco.2010.10.006 >.

1

Is Silence Golden? An Empirical Analysis of Firms that Stop Giving Quarterly Earnings Guidance in the post

Regulation-FD period

1. Introduction

On December 13, 2002, the Coca Cola Company announced that it would stop providing

quarterly and annual earnings-per-share guidance to stock analysts, stating that the company

hopes the move would focus investor attention on long-run performance. Shortly afterwards,

several other prominent firms such as AT&T and McDonalds made similar announcements

renouncing quarterly earnings guidance. In addition, recent surveys by the National Investors

Relations Institute (NIRI) suggest a trend toward firms discontinuing guidance or moving toward

providing annual guidance only.1 These changes in guidance practices coincide with calls from

prominent investors such as Warren Buffett (1996), analysts such as Candace Browning (2006),

head of global research at Merrill Lynch, the CFA Institute (Krehmeyer and Orsagh 2006) and

academics (Fuller and Jensen 2002, Jensen, Murphy and Wruck 2004) to encourage managers to

give up quarterly earnings guidance and hence avoid myopic managerial behavior caused by

attempts to meet the guided earnings number. However, critics allege that poor economic

conditions and the desire to de-emphasize weak performance drive these recent changes to

guidance policies (e.g., Harper 2003).

In this paper, we investigate 1) the factors that prompt firms to give up quarterly

guidance2, 2) the consequences of giving up guidance, and 3) changes in subsequent disclosure

1 A recent NIRI survey (reported in March 2005) reports that 71% of respondent firms provide some form of guidance (Thompson 2005), down from 77% in December 2003 (Thompson, 2003b). Moreover, the percentage of firms giving quarterly guidance has declined from 75% to 61% and the percent giving annual guidance only has increased – from 16% to 28%. 2 Although prior studies have examined characteristics of firms that, as of a point in time, give guidance versus those that do not (notably, Hutton 2005), no prior study, of which we are aware, has examined the determinants of the decision to either initiate or discontinue an earnings guidance policy. Anilowski et al. (2005) find that guidance has, in recent years, become more consistent, suggesting that the decision to provide earnings guidance appears to be a policy decision (as opposed to a decision that is made each quarter). As a result, a cross-sectional examination of

2

levels. To address these issues, we focus on firms that have stopped providing quarterly

guidance after the passage of Regulation FD (Reg FD). Focusing on the post FD period enables

us to control for the possibility that managers renounce guidance publicly but continue to give

private guidance to analysts.

Based on a detailed search of key words in press releases and conference call transcripts

from 10/1/2000 (the beginning of the FD regime) to 1/31/2006, we are able to identify 72 firms

that publicly announce their decision to stop providing quarterly guidance, and 24 firms that

publicly announce switching from providing quarterly earnings guidance to providing annual

guidance only. It is interesting to note that only a small number of firms have chosen to stop

guiding the market despite calls from prominent regulators and academics encouraging firms to

do so. The relatively small number of firms that have stopped providing earnings guidance

perhaps points to the high perceived costs of bucking the general trend of providing guidance.

We find that firms that stopped guidance have poor trailing 12-month stock return

performance. We do not find evidence that these firms have more long-horizon investors.

Together these results suggest that, on average, the primary driver behind the decision to stop

guidance is not to focus investors on the long-term, as many firms claim. We also find that firms

with lower institutional ownership and lower analyst following are more likely to give up

guidance – consistent with firms giving up guidance when demand for such guidance is lower.

However, the relation between analyst following and the decision to give up guidance is non-

linear – for firms with very high analyst following (in the top quartile of the distribution), analyst

following is not related to the decision to stop guidance, consistent with the idea that firms with

characteristics related to firms who do and do not give guidance in a particular quarter or year may not reveal the true underlying determinants of the decision to adopt such a disclosure policy (to the extent the characteristics have changed since the initial decision).

3

very high analyst following do not feel as much pressure to continue guidance in order to avoid

losing analyst following.

We document a significant negative market reaction to the firm’s announcement related

to renouncing guidance. This negative reaction could either 1) suggest that stopping guidance is

a signal about future performance; or 2) imply a revision in the cost of equity capital due to

changes in systematic risk. In additional analysis, we find evidence supporting primarily the first

explanation. A firm’s future performance is positively related to the announcement period return

– i.e., firms with worse future performance suffer a greater market decline when they announce

the decision to stop providing guidance. We also find that firms giving up guidance experience

an increase in systematic risk but we are unable to document an association between our

estimated change in systematic risk and the three-day market reaction to the stoppage

announcement.

We also find that the elimination of guidance results in stock prices reflecting earnings

news slower than when guidance is provided (i.e., that prices lead earnings less once guidance is

eliminated). Contrary to the beliefs of many proponents of earnings guidance, we do not find a

significant increase in overall volatility nor do we find a decrease in analyst following after the

firm stops providing guidance. We do, however, document a greater increase in analyst forecast

dispersion and a greater decrease in forecast accuracy following the decision to stop guiding for

our event firms relative to our control firms.

Finally, we find that firms disclose more information in their earnings announcement

press releases in the quarter after the decision to stop guidance relative to the quarter before.

This finding is consistent with firms substituting earnings guidance with greater qualitative

disclosures about the firm – a change that is supported by groups like the CFA Institute

4

(Krehmeyer and Orsagh 2006).3 However, these disclosures are not perfect substitutes for

earnings guidance as forecast accuracy decreases following the decision to stop guidance. Thus,

it does not appear that analysts are able to generate similar quality of information through the

analysis of these additional disclosures.

Our paper contributes to the literature on voluntary disclosures along several dimensions.

First, our study differs from prior disclosure studies in that we have identified a sample of firms

that announce a distinct material shift in their disclosure policy. The fact that we have identified

the date on which the firm announces this shift provides several advantages associated with

research design. First and foremost, a distinct event date allows us to more accurately document

the market reaction to this stoppage decision. Further, we are able to examine the association

between the market reaction and subsequent changes in systematic risk and future performance,

thereby providing further evidence on the association between disclosure and cost of equity

capital. In contrast, most prior studies use some measure of the level of disclosure (for example,

analyst ratings of corporate disclosure policies) and a cost of capital derived from equating

analysts’ forecasts of future earnings and current stock price. Studies in levels are, in general,

subject to the criticism of correlated omitted variables. Using a different methodology to assess

the impact of disclosure policy on cost of equity capital helps to triangulate prior results on this

relation.

On a related note, identifying the exact date of a change in disclosure policy allows us to

examine consequences related to stopping disclosure over relatively short windows surrounding

3 In July 2006, the CFA Institute Centre for Financial Market Integrity and the Business Roundtable Institute for Corporate Ethics co-sponsored the “Symposium Series on Short-Termism.” One of the recommendations arising out of the symposium was a reform of earnings guidance practices that called for 1) an end to the practice of providing quarterly earnings guidance and 2) support for corporate transitions to “higher quality, long-term, fundamental guidance practices, which will allow highly skilled analysts to differentiate themselves and the value they provide their clients.” (Krehmeyer and Orsagh 2006).

5

the disclosure policy change (such as changes in stock return volatility, informativeness of

earnings for stock returns, analyst following etc.). Prior studies often examine consequences of

disclosure changes over longer windows, which may be subject to confounding influences

unrelated to the change in disclosure levels (e.g., shifts in AIMR disclosure scores over a ten-

year period as in Healy et al. 1999).

Second, our study provides further insight into the motivation behind the recent upsurge

of firms discontinuing earnings guidance. Although many of these firms argue that guidance

forces a short-term orientation and impedes long-term value creation,4 our results suggest that

poor prior performance and lack of demand for guidance are the primary drivers behind the

decision to give up guidance. In particular, we are unable to document a positive association

between the decision to renounce guidance and the level of ownership by investors with longer

horizons and greater activism (public pension funds and block holder ownership). Moreover, if

guidance is truly a value-decreasing proposition on account of the short-term focus imposed on

managers, giving up guidance ought to be associated with positive announcement period returns.

However, we find an economically and statistically negative stock market reaction around the

guidance stoppage announcement.

Finally, our study should also be of interest to CFOs and investor relations professionals.

Recent surveys by the National Investor Relations Institute (NIRI) indicate that a significant

number of firms have considered discontinuing earnings guidance – 28% in a survey conducted

in February 2003 and 19% in a survey conducted in December 2003 (NIRI 2003). Moreover,

30% of survey respondents believe that if they stopped providing guidance, analyst coverage

would fall. Our study provides evidence on the consequences associated with the decision to

4 Cheng, Subramanyam and Zhang (2005) examine whether investment policies of regular guiders differ from those of sporadic guiders and find evidence suggesting that regular guiders under-invest in R&D (which they conclude is evidence that guidance forces a short-term orientation).

6

give up guidance. Contrary to many managers’ concerns, we do not find evidence of either a

decrease in analyst following or an increase in volatility around earnings announcements.

However, we do find an increase in analyst forecast dispersion and a decrease in analyst forecast

accuracy, despite the fact that firms, on average, provide additional disclosures in their press

releases. Thus, it appears analysts are unable to generate the same level of accuracy without

explicit earnings guidance.

The remainder of the paper is organized as follows. In the next section, we discuss our

sample and provide descriptive data. Section three presents our predictions and analysis of the

determinants of stopping guidance. In section four we examine the consequences associated with

discontinuing guidance. In section five we examine changes in firm’s disclosure levels following

the stoppage announcement. We conclude the paper in section six.

2. Background, sample selection, and descriptive data

Prior to the passage of Reg FD, firms could provide earnings guidance to the market

either indirectly through analysts or directly via publicly announced management forecasts.5 The

purpose of engaging in earnings guidance is, arguably, to keep expectations at a level that the

firm can either meet or exceed and, thereby, avoid any market penalties associated with missing

analysts’ expectations (Bartov et al., 2002; Skinner and Sloan 2002). Whatever the purpose, the

practice of providing earnings guidance is relatively widespread. Until recently, surveys by

NIRI have consistently reported that around 79% of respondents provide some form of earnings

guidance (Thompson 2003a and 2003b). However, in their most recent survey, the proportion of

guiders has declined to 71% (Thompson 2005). Nevertheless, the practice of providing guidance

5 One of the most common ways managers provided guidance indirectly through analysts was by “reviewing” analysts’ earnings models (Hutton 2005).

7

is still relatively widespread, which likely makes the decision to stop providing guidance

difficult.

We focus our study on the period following the passage of Reg FD. Thus, we define

earnings guidance as the practice of providing regular, quantitative forecasts of upcoming earnings

either via press releases or company-sponsored conference calls. To identify firms that have

changed their disclosure practice from providing quarterly earnings guidance to either providing

no guidance or providing annual guidance only, we conduct a detailed search of company press

releases and conference call transcripts between October 1, 2000 and January 31, 2006. We search

company press releases on PR Newswire and BusinessWire (via Lexis/Nexis) and conference call

transcripts on the Fair Disclosure Wire (via Factiva).

Our main search string was “(earnings or income or loss) and ((guidance or expectation

or forecast or outlook) w/5 (no longer or stop or discontinue or will not provide or will not

give)).” We then read the press release or conference call transcript to determine whether the

statement did indeed refer to a change in the company’s policy regarding the dissemination of

earnings guidance.6 Through this process, we identify 72 firms that publicly announce their

decision to stop providing quarterly guidance altogether and 24 firms that switch from providing

quarterly earnings guidance to providing annual guidance only.7 For better statistical power, we

6 Often the search string identified firms that indicate that previously issued guidance “no longer applies” or that the company has temporarily stopped guidance until “visibility improves”. We do not include the latter group in the sample as we are more interested in firms who have indicated a commitment to a new disclosure policy. 7 To determine whether these firms live up to their promise to give up guidance, we check whether these firms appear on the CIG database subsequent to their stop announcement. Twenty-three firms appear on the CIG database at least once after their announcement. Upon further investigation we discovered that in 10 cases the observations on CIG were pre-announcements (announcements of forthcoming earnings that generally occur within 2-3 weeks of the actual earnings announcement) not earnings guidance, 4 cases were “switchers” who provided annual guidance in the 4th quarter which was coded as quarterly guidance by CIG, 3 cases were firms who made one-time exceptions to their policy, 1 case was due to a lag in the initiation of their new policy (e.g., they announced the decision to stop at the end of the current year), and 1 case was an instance of “qualitative” guidance (i.e., “(the company’s) return to profitability will last through the first half of 2004”). The remaining four cases represent firms that appear to have re-started guidance after announcing the decision to stop. Our main results are not affected if we eliminate these four firms from our tests.

8

combine these two groups of firms as both sets of firms have stopped providing regular quarterly

guidance.8

An alternative approach to identifying a set of firms that stop providing regular quarterly

guidance would be to identify firms who stop appearing on First Call’s Company-Issued

Guidance (CIG) database. There are two advantages with our approach. First, searching for

firms’ announcements of the decision to stop providing guidance allows us to test the market

reaction to the announcement of this decision. Using the CIG database would not provide a

specific date associated with the decision to stop guiding. Second, using the CIG database

requires us to infer a company’s disclosure policy (and any change therein) based on some

pattern of appearances of management forecasts in the database (e.g., eight quarters in a row

followed by no forecasts for eight quarters). This is particularly difficult given that, pre-Reg FD,

firms could provide guidance privately to analysts and therefore not appear on the CIG database.

This fact leaves a relatively short time-frame to both establish a pattern of guidance as well as

allow sufficient time to ensure the absence of guidance. Because of this data limitation, inferring

from CIG database when a firm stops guidance depends completely on a researcher’s ad hoc

definitions of both how many quarters of appearance on CIG is needed to establish a guidance

pattern and how many quarters of non-appearance on CIG is long enough to constitute a ‘policy

change’ of stopping guidance.9 The downside to our approach is that we might miss firms that

8 We compared the “stoppers” and “switchers” along the characteristics discussed in the next section. No differences in medians are significant at traditional levels. We also examine the 3-day market reaction surrounding the announcement of the policy change and do not find significant differences between the two groups in terms of either raw returns or cumulative abnormal returns. Thus, the two groups do not appear significantly different from another and combining the two groups seems reasonable. 9 For example, Houston, Lev, and Tucker (2006) rely on firms’ non-appearance on CIG database to infer guidance stoppage. A more serious measurement error can occur with this approach if the CIG database is incomplete (i.e., not all guidance provided by a firm is reported on the database). Though the extent of CIG’s omission of management forecasts is unknown, we have noticed multiple instances where a firm issues a press release containing management forecasts but is not captured by CIG.

9

stop giving guidance but do not publicly announce their intention to do so. Thus, our

conclusions apply only to firms who publicly announce the decision to stop providing guidance.

Figure 1 provides a distribution of announcement dates by month from the first

announcement date. There are several interesting facts that are apparent from our sample. First,

the number of firms that have publicly announced the decision to stop providing quarterly

earnings guidance is relatively small.10 This fact is, perhaps, indicative of the difficulty

associated with changing accepted business practice. Anecdotal evidence suggests that

companies can indeed face powerful opposition to stopping guidance, as analysts’ compensation

and CEOs reputation in recent years all depend to a greater extent on the rapid delivery of

information from the firms.11 Second, there were relatively few firms who stop giving guidance

before January 2003. It is interesting to note that Coke’s decision to stop providing guidance

occurred in December 2002, immediately before the peak in announcements. Coke’s

announcement was highly publicized and as a large, well-respected company, likely provided

companies with support for their decision. At least 13 firms explicitly cite “consistency with

practice” as their reason for changing their disclosure policy. For example:

• “Datascope also announced that, along with other leading companies, it will discontinue its practice of offering earnings forecasts.” (Datascope press release, 1/23/2003).

• “As many companies have done recently, we have examined our practice of providing forward-looking guidance and have made the decision to no longer provide

10 The small sample size is consistent with prior NIRI survey results. In an April 2003 survey, 22% of the 609 respondents indicated that they do not provide guidance and, of these, 22% stopped within the last 12 months. This suggests approximately 29 respondent firms stopped giving guidance after April 2002 (609 × 22% × 22%). The response rate on the survey was 20%. Thus, if the rates of stoppage are representative of the overall rates in the NIRI membership, the number of firms stopping guidance from the NIRI membership as a whole would be approximately 145 (29 ÷ 20%). However, Hutton (2005) finds evidence consistent with respondent firms having more active investor relations groups; thus, it is possible the rate of stoppers is less within the population of non-respondents (since firms with more active investor relations groups are more likely to have a policy of giving guidance and a firm can not stop giving guidance if it did not initially have a policy of giving guidance). In addition, there may be firms that stop giving guidance but do not announce this decision publicly. 11 See “The wrong focus? How the race to meet targets can throw corporate America off course,” Financial Times (7/24/2006).

10

forward-looking guidance. We believe that our decision is consistent with emerging corporate disclosure trends, seen in financial markets today.” (Frank Terence, Brightpoint CFO, Conference Call Transcript, 5/1/2003).

• “In light of changing events and emerging corporate disclosure trends seen in financial markets today, the Company has examined its practice of providing forward-looking financial guidance and has made the decision to no longer provide forward-looking financial guidance.” (MediaBay, Inc. press release, 6/20/2003).

Untabulated descriptive data on the reasons given by firms for stopping the use of

quarterly guidance also reveals the following: approximately 28% of firms provide no

explanation for their change. Of those who do provide reasons, “difficulty in

predicting/uncertainty” (25%) and “desire to focus on the long-term” (23%) are the most

common reasons given.

Table 1 details the sample attrition due to data requirements for our subsequent tests. We

conduct three groups of tests: 1) tests of the determinants of the decision to stop providing

earnings guidance (section 3), 2) tests of the market reaction to the announcement of the stop

guidance decision (section 4), and 3) tests of the consequences resulting from the decision to stop

providing earnings guidance (section 4). The usable sample for each of these three tests varies

from 53 to 75. In some of our market reaction tests, we control for revisions in analysts’

forecasts of future earnings, which reduces our sample size to 53 (from 63). In our consequences

tests, we examine changes in various factors (e.g., analyst following, volatility, volume) before

and after the decision to stop providing earnings guidance. Thus, we lose firms who announced

in late 2005 and early 2006 (due to lack of post-decision data at the time of drafting this paper).

In the next section, we discuss our hypotheses and tests related to the determinants of

stopping earnings guidance.

3. Determinants of giving up guidance

3.1 Empirical Predictions

11

3.1.1 Focusing on the long-term or hiding poor performance?

Much of the current debate over earnings guidance centers around whether this practice

engenders managers with a short-term focus.12 Moreover, in justifying their decision to give up

earnings guidance, many firms cite a desire to focus on long-term performance as the reason

behind their decision to stop providing quarterly guidance. For example:

• “Following a series of discussions with our Board of Directors over the past year, our management team has established a policy of not providing quarterly or annual earnings guidance…we believe that establishing short-term guidance prevents a more meaningful focus on the strategic initiatives that a Company is taking to build its business and succeed over the long-run…Our share owners are best served by this because we should not run our business based on short-term ‘expectations.’ We are managing this business for the long-term.” (Coca cola, press release, 12/13/2002).

• “Management of the Company believes that the focus placed on achieving short-term earnings estimates detracts from the Company’s strategy to create long-term value for its shareholders.” (DST Systems Inc, earnings release, 1/27/2004).

• “Following the recommendation of our board of directors, our management team will implement this policy to highlight the benefits of our strategy over the long term to employees and shareholders. The provision of revenue and earnings guidance encourages a short-term outlook which, in our view, is not in the best interests of our company or our shareholders.” (Scientific Games, earnings release, 2/26/2004).

However, critics have suggested that the true motivation behind these firms’ decision to

stop providing quarterly guidance is poor performance (Harper 2003).13 In addition, prior

studies have generally found a positive association between performance and disclosure (Lang

and Lundholm 1993; Miller 2002), suggesting that poor performance may at least partially

explain a firm’s decision to stop providing quarterly earnings guidance.14

12 See “The wrong focus? How the race to meet targets can throw corporate America off course,” Financial Times (7/24/2006). 13 In addition, The Boston Globe columnist Charles Stein writes, “Look at the list of companies that have stopped issuing guidance and ask yourself: What do they have in common? The answer: They all have performed poorly, not just recently but for many years.” (“Misguided Reform” The Boston Globe, 2/9/2003, C2). 14 Although prior studies have documented a positive relation between performance and disclosure, it is possible this relation would not exist in this setting because the decision to give up guidance represents a change in a company’s disclosure policy – one that involves a commitment to future actions (or rather the lack of future actions). It is possible that changes in performance, particularly if transient in nature, would not provide a strong enough motivation for a company to change its disclosure policy.

12

If the primary reason behind giving up guidance is the desire to focus on long-term

performance, we would not expect a relation between past performance and the decision to stop

guidance. In contrast, if the skeptics are correct in their assessment of the reason behind the

decision, firms with poorer prior performance are more likely to give up guidance.

We proxy for poor performance in three ways: 1) stock price performance in the recent

past, measured as market-adjusted buy-and-hold returns for the 12 months ending in month -1

(BH_1) and month -13 (BH_2), where month 0 is the event month; 2) the percentage of losses in

the eight quarters proceeding the event quarter (PLOSS), where loss is defined as net income less

than zero; 3) the percentage of quarters in the eight quarters proceeding the event quarter in which

the firm meets or exceeds analyst consensus forecasts (PMBAF).15

The desire to focus on long-term performance is more likely to arise in companies whose

primary shareholders have longer investment horizons and who take an active corporate

governance role – exercising influence on management decisions. If focusing on long-term

performance is truly a driving factor behind the decision to stop providing earnings guidance, we

would expect firms who stop guidance to have shareholders with longer investment horizons and

who take more active roles in corporate governance.

We focus on two types of shareholders with these qualities. First, public pension plans

have often played an active role in corporate governance (Del Guercio and Hawkins 1999) and are

also generally believed to be long-term investors (Cremers and Nair 2004). We measure pension

plan ownership as the percentage of shares held by the 18 largest public pension funds (PO).16

Second, blockholders (shareholders with greater than 5% ownership in a firm) are also believed to

be active monitors of the firm (Cremers and Nair 2004) because they generally have a longer

15 See notes to Table 2 for exact variable definitions. 16 We thank K. J. Martijn Cremers and Vinay B. Nair for sharing their pension fund data with us.

13

investment horizon. We measure the percentage of shares held by blockholders (BO) as reported

on Compact Disclosure.

3.1.2 Usefulness of earnings guidance

Dye (2001) argues that managers have incentives to make voluntary accounting

disclosures which investors find useful in assessing firm value. Prior research has generally

found that firms with low value relevance of earnings are more likely to provide voluntary

supplemental disclosures such as balance sheet information (Chen, Defond and Park 2002) or

conference calls (Tasker 1998). The argument in these studies is that when earnings are a poor

indicator of future cash flows, the market demands additional information to assess firm value.

In our setting, however, the supplemental disclosures or guidance is about earnings itself.

Hutton (2005) suggests that the market would demand more earnings guidance when earnings is

a better indicator of firm-value i.e., more value relevant. Hence, we predict that firms with less

value relevant earnings are more likely to give up guidance.

We measure value relevance of earnings (VALREL) as negative one times the squared

residual from a regression of annual, market-adjusted returns on 1) net income scaled by

beginning market-value of equity, 2) a loss dummy variable, 3) an interaction of net income and

the loss dummy variable, and 4) the change in net income (as in Ashbaugh et al. 2005). Thus,

higher values represent more value relevant earnings.

Earnings guidance is also more useful when firm performance is variable and harder to

predict based on mandatory accounting disclosures. In other words, if market participants can

easily predict future earnings based on mandatory disclosures, additional voluntary disclosures

are less useful. Prior research supports this conjecture: analyst ratings of firms’ disclosures are

higher (Lang and Lundholm 1993) and firms are more likely to include balance sheet data with

14

their press releases (Chen et al. 2002) when returns are more volatile. This suggests that firms

will be more likely to discontinue guidance when firm performance is less variable and easier to

predict.

Our proxy for the volatility of firm performance is based on stock returns – specifically, we

measure the standard deviation of daily raw returns (STDRET) during the 252 days prior to the

event date. Our proxy for earnings predictability (PRED) is negative one times the root mean

squared error ( )ˆ(2jvσ from a firm’s AR1 model of regressing seasonally adjusted quarterly net

income on a lagged version of such changes (Francis et al. 2004). Thus, higher values of PRED

represent more predictable earnings.17

3.1.3 External demand for guidance

The primary beneficiaries of a firm’s earnings guidance are financial analysts and

institutional investors. Analysts can use firm-provided guidance to formulate and validate earnings

forecasts and as a result, firms likely face pressure from analysts to provide guidance. Similarly,

the buy-side can use guidance to evaluate their investment decisions. In a recent NIRI survey, 98

percent of respondents believe analysts want earnings guidance and 27 percent of respondents

believed analyst coverage would drop if the firm stopped providing guidance. Thus, firms with

more analyst coverage and greater institutional investor ownership likely face greater pressure to

provide guidance and are less likely to discontinue the practice.

On the other hand, if a firm is prominent in its industry it will likely garner a large analyst

following regardless of whether the firm provides earnings guidance. Thus, it is likely that the 17 Note that in this hypothesis “predictability” refers to the market’s ability to forecast future earnings based on mandatory disclosures (e.g., historical earnings). When this is easily done, management-provided guidance provides little value, making it more likely for the firm to give up guidance. “Predictability” here does not refer to management’s ability to predict earnings using their private information. It is likely that the harder it is for management to predict earnings, the more likely they are to give up guidance. This effect is more likely captured by our variable PMBAF, the percentage of quarters in which the firm met or exceeded analysts’ expectations in the past.

15

relation between analyst following and the decision to stop providing guidance is not linear – firms

with very large analyst following are likely less concerned about losing analyst coverage.

We measure analyst following (AF) as the number of analysts covering the firm at the

beginning of the event quarter as per IBES. If a firm is not listed on IBES, we code AF as zero.

We also specify a dummy variable (HIAF) to indicate analyst following in the fourth quartile. In

our main analysis, we allow the coefficient on AF to vary between the first three quartiles and the

fourth quartile. The percent of institutional ownership (PINST) is measured prior to the

announcement date as reported on Compact Disclosure.

3.1.4 Litigation risk

Litigation fears can reduce incentives to provide disclosure, particularly disclosures of

forward-looking information (such as earnings guidance). A manager may fear the legal system

would penalize forecasts (made in good faith) that are not met because it cannot effectively

distinguish between unexpected forecast errors due to chance from those due to deliberate

management bias (Healy and Palepu 2001). Francis, Philbrick and Schipper (1994) document

that for a large sample of class action securities lawsuits between 1988 and 1992, over 80% are

based on earnings.

In an interview we conducted with a financial executive of a firm that gave up earnings

guidance, the executive stated “class action lawyers find it easier to sue the firm if it is unable to

meet the EPS numbers promised in earnings guidance.” Thus, practitioners are potentially

concerned about being sued for the accuracy of its forecasts under rule 10(b)5. Furthermore,

companies are beginning to require audit committees to review any earnings guidance they give.

That practice raises the possibility that audit committees could also be sued for providing

16

earnings guidance, thus adding further to the litigation risk (Morgan, 2003). Hence, we expect

firms with greater litigation risk to be more likely to give up earnings guidance.

Our proxy for litigation risk is based on the Rogers and Stocken (2005) litigation

probability model. We use the reported coefficients from their model to compute predicted

values for the observations in our sample (LIT).

3.2 Empirical design and results

3.2.1 Control sample

To examine the characteristics of firms that give up earnings guidance, we identify a

control sample of firms that have not discontinued guidance. We identify all firms available on

the First Call CIG database that provided a quarterly management forecast within +/- 90 days of

each event firm’s announcement date and who also issued at least one quarterly forecast in the

previous and subsequent quarters (quarters q-1 and q+1).18 A control firm can only appear once

in the sample but can serve as a control for more than one event firm. In addition, we allow

multiple control firms for each event firm. This process results in 901 control firms. For the

measurement of variables, the ‘event’ date for control firms is the date the management forecast

is issued in quarter q.

3.2.2 Univariate statistics

Table 2, Panel A presents descriptive statistics for the variables used in our analyses,

partitioned by event and control firms. We include three control variables in our analysis: 1)

LNMV, the natural log of the market value of equity as of the fiscal quarter end preceding the

event date; 2) MB, the natural log of the market-to-book ratio at the beginning of the quarter, and

18 Ideally, our comparison group should be regular guidance givers who continue to give guidance. Requiring our control sample to have forecasts in quarters q-1 and q+1 provides some assurance that these firms have a pattern of giving guidance in the past and continue to give guidance in the future. The mean number of quarterly forecasts made by these firms through quarter q-1 is 17.58 and over 70% of the firms have made more than 10 quarterly forecasts.

17

3) LNCT, the natural log of (1+CT), where CT is the number of management quarterly EPS

forecasts made through quarter t-1 (as reported on the CIG database). Prior research has shown

both size and growth to be related to disclosure practices. In addition, firms with a longer prior

history of providing earnings guidance likely find it more difficult to abandon the practice.

We report t-tests and Wilcoxon z-tests for differences in means and medians between the

two sets of firms. Several differences are statistically significant. Firms stopping guidance

exhibit poorer recent performance than control firms – lower stock returns over the past year

(BH_1), a higher percent of prior quarterly losses (PLOSS), and a lower percent of prior quarters

where firms met or exceeded analyst forecasts (PMBAF). Also, firms who stop guidance have

fewer analysts following the firm (AF) and a lower percent of institutional investors (PINST).

We also find that firms that give up guidance have a shorter history of giving guidance (LNCT).

When comparing means, the probability of litigation (LIT) is significantly smaller for firms that

stop giving guidance (contrary to our predictions) but is significantly larger when comparing

medians (consistent with our predictions). We also find some differences that are significant in

means but not in medians (PRED) and vice versa (VALREL and MB). Thus, it appears that

extreme observations drive some of these differences. We address the issue of extreme

observations in the next section by running our regression using ranked values.

In addition, there are several variables that are highly correlated with each other. Table 2,

Panel B presents a correlation matrix of the variables used in this study. Firm size (LNMV) is

highly correlated with volatility of returns (STDret) (Pearson correlation coefficient ρ=-0.46) and

analyst following (ρ= 0.67). Poor performance (as measured by prior losses, PLOSS) is also

highly correlated with volatility of returns (ρ=0.59). Given the high degree of correlation

between many of our variables, our primary tests rely on multivariate relations that consider

18

together all factors that we hypothesize to be associated with the likelihood of guidance

stoppage. We discuss these tests next.

3.2.3 Logistic regression

We test our hypotheses by estimating the coefficients in the following logistic regression:

0 1 2 3 4 5 6

7 8 9 10 11 12 13

14 15 16 17 18

_1 _ 2 ret

b b

p p

STOP BH BH PLOSS PMBAF VALREL STDPRED AF HIAF AF PINST LIT D D BOD D PO LNMV MB LNCT

β β β β β β ββ β β β β β ββ β β β β ε

= + + + + + ++ + + × + + + + ×

+ + × + + + +

(1)

STOP is a dummy variable equal to one for event firms and zero for control firms.19

Because we do not have pension fund and blockholder ownership data for our entire sample of

firms, we code the ownership variables (BO and PO) as zero for firms without available data and

interact these variables with a dummy variable (Db and Dp) that is equal to one for firms with

data available and zero otherwise. This specification, called modified zero-order regression

(Greene 1993), addresses the selection bias related to coverage of the two data sources while

maintaining sample size.

We estimate our regression in both raw values as well as ranked values for the independent

variables, as the univariate comparisons suggest the presence of extreme values for some variables.

Table 2, Panel C presents the results from estimating equation (1).

Firm performance appears to be a significant determinant of the decision to stop

guidance, consistent with our first hypothesis. Firms who stop guiding have lower lagged one-

year buy-and-hold returns (p’s <0.01 for both raw and ranked values). These firms also show

19 As discussed in the previous section, we do not find significant differences between the “stoppers” and the “switchers” and therefore, combine these two groups. If we instead run an ordered logistic regression coding control firms as 0, switchers as 1, and stoppers as 2, we find similar significance levels on the coefficients. However, Greene (1997, p. 929) points out that in ordered logistic regressions, the sign of the change in probability associated with changes in the independent variables are unambiguous only for the first and last ordered group; the direction of the effect for the middle group depends on the shape of the density function. Given the difficulty of interpreting coefficients from these models and the similarity between the two groups, we do not report the results from the ordered logistic regression.

19

some evidence of poorer prior earnings performance – using raw values, they exhibit a lower

percent of prior quarters that met or exceeded analysts’ expectations (p=0.07) and using ranked

values, they exhibit a higher percent of prior quarterly losses (p=0.06). In contrast, we do not

find a significant relation between shareholder horizon and the decision to stop providing

earnings guidance – coefficients on both Db×PO and Dp×BO are insignificant. The combined

evidence suggests that poor performance is the primary driver behind the decision to stop

providing earnings guidance – contrary to claims that firms have stopped guidance in order to

focus shareholder attention on long-run performance.

We also find evidence that firms with lower demand for guidance are more likely to give

up guidance. Specifically, firms with lower analyst coverage and lower percent of institutional

investors are more likely to give up guidance – the coefficients on AF and PINST are statistically

negative using both raw and ranked values. However, this relation is attenuated for firms with

very high analyst following – the coefficient on HIAF×AF is significantly positive. This

evidence suggests that firms with very high analyst following are less concerned about analysts’

demands for earnings guidance.

We also find some evidence that firms are more likely to stop guidance when their

earnings guidance is less useful – firms who stop guiding have lower value relevance of earnings

and less volatile performance but these variables are only significant in the ranked regression.

Similarly, there is some evidence that firms with higher litigation risk are more likely to give up

guidance but the coefficient on LIT is only significant in the ranked regression.

Turning to the control variables, firms with longer histories of providing earnings

guidance are less likely to give up guidance. Using raw values, we also find that larger firms are

more likely to give up guidance but this relation does not exist using ranked values.

20

Overall, our results most strongly support the hypotheses that poor performance and

external demand for guidance drive the decision to stop providing earnings guidance. Thus,

while many firms provide virtuous reasons for no longer providing guidance (e.g., desire to focus

on the long-term), we do not find that firms who renounce guidance have greater ownership by

long-horizon, active investors.

4. Consequences of giving up guidance

4.1 Market reaction to announcement of stoppage

In this section, we investigate whether firms that stop providing earnings guidance

experience abnormal returns when they announce their intentions. On the one hand, firms often

claim that the reason for giving up guidance is to avoid focusing investor as well as managerial

attention on the short-run performance of the company. If guidance is truly a value-decreasing

proposition and the market realizes this, announcement period returns should be positive.

However, prior research suggests a negative relation between disclosure and cost of capital

(Botosan 1997; Healy, Hutton, and Palepu, 1999), which would suggest that a firm stopping

guidance would experience a negative stock price reaction.20

One potential complication is that many firms (54 out of 73) announce the decision to

stop guidance in conjunction with their announcement of quarterly earnings. Thus, the

announcement period return will capture the effect of both the stoppage announcement as well as

the earnings announcement.21

20 Botosan and Plumlee (2002), however, find a positive relation between cost of capital and disclosure scores related to “other publications,” which include “quarterly and other published information not required.” Assuming guidance falls in this category of disclosure, their findings would suggest that guidance increases cost of capital and firms who stop guiding should experience a positive market reaction (due to a lower cost of capital). 21 Some firms’ earnings announcements also include guidance (generally downward) for the upcoming quarter or year (with a statement indicating that the firm will not be providing guidance going forward). In this case, a negative market reaction could be attributed to revised expectations about future earnings. In subsequent analyses, we examine whether the market reaction is related to revisions in future earnings expectations.

21

Table 3, Panel A provides descriptive statistics on the three-day (centered on the

announcement date) cumulative abnormal returns (CAR) as well as cumulative raw returns

(RET), for firms that announce the stoppage in conjunction with their quarterly earnings

announcement (Dea=1) and those that announce the stoppage independently (Dea=0). Both sets

of firms experience a statistically significant negative three-day abnormal return as well as a

three-day raw return. More importantly, the three-day return is not statistically different between

the two groups. Overall, the three-day return is negative for roughly 66% of the sample.

To formally control for the news in the earnings announcement, we run the following

regression

1, 1 1 2 ( )ea eaCAR D D UEα β β ε− + = + + × + (2)

In this model, α captures the average three-day market reaction to the stoppage announcement,

Dea captures any differential market reaction for the group announcing earnings in conjunction

with the stoppage announcement, and Dea × UE captures the market response to the earnings

surprise. The results are reported in Table 3, Panel B. The intercept indicates that the average

market reaction, after controlling for the surprise in earnings, is a statistically significant negative

return of –4.8% (p<0.01).22

The negative market reaction indicates that the market penalizes firms for stopping

guidance, suggesting that investors value earnings guidance. This result is consistent with prior

research that finds a negative relation between the level of voluntary disclosure and the cost of

equity capital. If disclosure reduces the cost of equity capital then a firm’s decision to eliminate

22 Given the significant amount of clustering in announcement dates (as shown in Figure 1), we also ran firm specific regressions and tested the significance of the average intercept term. Specifically, we ran the following regression: RET it = αi + βi MRETt + γi EVENTit + εit, where RETit is firm i’s daily stock return; MRET is the value-weighted daily market return, including dividends; EVENTi is an indicator variable equal to 1 for firm i’s stoppage announcement, and 0 otherwise, and t denotes each of the 252 trading days in the calendar year of the stoppage announcement. The mean coefficient γi across the 73 event firms is −0.012 and is statistically significant at p <0.01. Thus, our results do not appear to be significantly affected by cross-sectional correlation.

22

earnings guidance, one form of disclosure, would increase the firm’s cost of capital and result in

a negative price reaction at the time of the announcement.

To further investigate this possibility, we examine changes in the systematic risk of the

firm before and after the announcement to stop providing guidance. We estimate the market

model using monthly returns including a dummy variable, POST, which is equal to one for

months subsequent to the stop guidance announcement. Specifically, we estimate the following

regression for each event and control firm:

[ ] [ ]it ft post i i mt ft post i mt ft itR R POST R R POST R Rα α β β ε− = + + − + × − + (3)

Ri,t is the monthly stock return for firm i, Rf,t is the monthly risk-free rate, and Rm,t is the monthly

return on the NYSE-AMEX-NASDAQ value-weighted market portfolio. POST is an indicator

variable coded as 1 for the months after the stop-guidance announcement, and 0 otherwise. We

include months up to month -60 prior to stopping and all available months after stopping,

eliminating firms with less than six months of returns in the post stoppage period. 23

We report the mean and median values of βPOST for the event and control samples in

Table 4, Panel A. The mean and median values of βPOST are greater than zero (p’s < 0.01, one-

tailed) suggesting an increase in systematic risk for event firms. However, βPOST is also

significantly greater than zero for the control group, though the difference between the two

23 This design is similar to that used to estimate changes in risk surrounding the announcement of dividend changes (Grullon, Michaely, and Swaminathan, 2002) and open market share repurchase programs (Grullon and Michaely 2004). We do not use the Fama and French (1993) three-factor model for two reasons. First, estimating factor loadings on SMB and HML on a firm-specific basis is empirically noisy but there is a long tradition of estimating one factor CAPM models on a firm-specific basis in accounting and finance research. Note that Fama and French (1993) estimate three factor loadings on a portfolio basis. Second, the three factor Fama French model in place of the one factor CAPM model, even on a portfolio basis, runs into an asset spanning problem in that the loading on the CAPM market factor can be potentially distributed over all three factors of the Fama-French model. Hence, the researcher can potentially encounter increases in one or two factor loadings (e.g., loading on the market factor and SMB) and a decrease in another factor loading (say HML), thus rendering interpretation difficult. In contrast, the direction of change in the loading on one of the three factors has a relatively unambiguous interpretation.

23

groups is significant: for event firms beta increases more after the guidance stop announcement

than for control firms.

However, because firms may begin with different levels of systematic risk, we also consider

the percent change in CAPM beta, CBETA, defined as βPOST ÷ βi. These results are presented in the

last two columns of Panel A, Table 4. Event firms show a highly significant percentage increase in

CAPM beta using both means and medians. Control firms show a significant increase in medians

but a decrease in means. We also find a significantly greater CBETA for event firms relative to

control firms when comparing both means (t=1.85) and medians (z=2.41). Thus, we have

preliminary evidence that the negative market reaction to the stop guidance announcement is, at least

partially, related to a shift in risk.

An alternative explanation for the decline in market price is that the market interprets the

decision to stop guiding as a signal about future performance – in other words, the market revises

downward its expectations about future cash flows of firms that renounce guidance. To provide

some evidence on this possibility, we run the regression in Equation 2, controlling for changes in

expectations of future earnings (ΔFEARN). Specifically, we calculate the revision in analysts’

earnings forecasts for quarter q+1 made after the stop guidance announcement. If the negative

market reaction is due to revisions in future expected cash flows, the coefficient on ΔFEARN should

be positive – that is, firms with greater declines in expected future performance should have more

negative market reactions at the time of announcement.

The results of this analysis are presented in the first column of Panel B of Table 4. The

coefficient on ΔFEARN is positive and statistically significant (t=3.83, p<0.01), indicating that

firms with more negative revisions in expected future performance are penalized with a more

negative market reaction at the time of the announcement. However, the intercept is still

24

significantly negative (t=-2.14, p= 0.05), suggesting that changes in risk may also be an

incremental explanation for the negative reaction.24

We further examine the role of changes in cost of capital in explaining the negative

reaction by including the decile values of our measure of changes in systematic risk, RCBETA,

in equation 2.25 If the negative reaction is due to expected changes in systematic risk, the

coefficient on RCBETA should be negative. Results of this analysis are presented in column 2

of Table 4, Panel B. The coefficient on RCBETA is insignificant. If we include both ΔFEARN

and RCBETA into the model simultaneously (column 3), the coefficient on ΔFEARN remains

positive and significant and the coefficient on RCBETA remains insignificant.

Finally, we run the above analysis after including proxies for the following 1) prior

performance, 2) demand for information, 3) the company-reported reason for stopping, and 4)

firm size. Our prior analysis finds that prior performance is an important determinant of the

decision to stop providing earnings guidance. Thus, we include prior returns (BH_1) and the

percent of prior quarters in which the firm met or exceeded analyst expectations (PMBAF) to

control for any systematic difference between firms with prior poor performance and

announcement period returns. Our determinant tests also find that demand for information is

another driver of a firm’s decision to stop giving guidance, thus we include analyst following

(AF) and institutional holdings (PINST) in the model. We also include indicator variables

designating firms whose self-reported reason for stopping guidance was due to 1) the

unpredictability of performance (LOPRED) and 2) the desire to focus on the long-run

performance of the company (LTFOCUS). It is possible the market reaction is related to the

24 It is possible that the reaction is due to revisions in longer run performance. However, the availability of long-term forecasts in IBES is limited. Given our relatively small sample of event firms, we are unable to examine changes in long-term forecasts. 25 We use decile values of CBETA to abstract away from large outlier changes in CBETA. Our inferences remain unchanged when we use actual values of CBETA in the analyses.

25

reason provided by the company. Finally, we include the log of market value of equity to control

for firm size.

Column 4 of Table 4, Panel B reports the results of adding only the prior performance

and demand for information variables as additional controls and Column 5 reports the results of

including all the above variables. Similar to our prior results, revisions in expectations about

future performance (ΔFEARN) are positively associated with announcement period returns while

changes in systematic risk (RCBETA) are not. In terms of the control variables, PMBAF is

positive and significant, indicating that firms with a history of meeting or beating analysts’

expectations fare better, in terms of market reactions, when announcing their decision to stop

providing guidance. Prior returns, analyst following and institutional ownership do not appear to

influence the market’s reaction to the announcement. However, it is interesting to note that the

coefficient on LTFOCUS is positive and significant (t=2.27, p<0.03, one-tailed), suggesting that

firms who state that their reason for giving up guidance is to focus investor attention on the long-

term are penalized less severely.26

Overall, our results suggest that the negative reaction is related primarily to revisions in

expected future performance. While we find evidence of an increase in systematic risk following

the decision to give up guidance, this change is not correlated with market returns at the time of

the announcement. One possibility is that firm-specific measures of changes in systematic risk

are too noisy, resulting in a low power test.

4.2 Other consequences

Our last set of tests investigates other potential consequences of giving up guidance. We

group these consequences into three categories: 1) effects on the timing of earnings news, 2)

26 Untabulated results show that proxies for long-horizon investors (block and pension fund ownership) are unrelated to the cross-sectional distribution of CARs.

26

effects on volatility and trading volume, 3) and effects on analyst-related factors. For all these

tests, we designate the quarter prior to the stoppage announcement as the “pre-stoppage” quarter

and designate the quarter following the announcement as the “post-stoppage” quarter. We test

for differences between the pre- and post-stoppage quarters for both the event and control

samples. Thus, our basic research design is a “control group design with pre-test and post-test”,

which has fewer validity threats than a “pre-test only” design (Cook and Campbell, 1979). In

addition, we include the Inverse Mills Ratio (IMR) obtained from the determinants test to control

for potential self-selection bias inherent in our research setting.

4.2.1 Timing of earnings news

We first examine the timing of earnings news to the market. One potential benefit of

earnings guidance is that the market receives information about upcoming earnings surprises

earlier in the quarter. When a firm stops giving guidance, less (more) of the earnings

information will be incorporated in price before (during) the earnings announcement period. We

thus examine changes in the relation between 1) total earnings news released during the quarter

and returns prior to the earnings announcement date (EAD) (equation 4a) and 2) total earnings

news and returns during the earnings announcement window (equation 4b):

[ 2, 2], 0 1 2 , 3 , 4 , 1 ,( )j q j j q j j q j q j qCAR POST UE POST UE IMRα α α α α ε+ −

−= + + + × + + (4a)

[ 1, 1], , 0 1 2 , 3 , 4 , 1 ,( )j q EAD j j q j j q j q j qCAR POST UE POST UE IMRα α α α α ε− +

−= + + + × + + (4b)

In both equations, UE is the seasonally differenced EPS (Compustat quarterly #19),

scaled by stock price at the beginning of the pre-stoppage quarter, and POST is a dummy

variable equal to 1 for the post-stoppage quarter. We measure UE relative to the same quarter in

27

the prior year as we are attempting to capture total earnings news released during the quarter.27

IMR is the Inverse Mills Ratio obtained from the determinant test as specified in Equation (1).

We include IMR to control for any potential self-selection bias arising from omitted firm

characteristics that are correlated with the decision to stop giving guidance.

In equation 4a, [ 2, 2],j qCAR + − is the cumulative abnormal return from +2 days following the

prior quarter’s earnings announcement to -2 days before the current quarter’s earnings

announcement date (i.e., the earnings announcement of UE). See Figure 2 for a depiction of the

variable measurement for this test. Thus, in this equation the coefficient on α2 captures the

amount of the total earnings news that is revealed prior to the earnings announcement date and

α3 captures the change in this anticipation after the firm stops guiding. If, after stopping

guidance, less of the total earnings news is revealed prior to the earnings announcement, then the

coefficient on α3 should be negative.

In equation 4b, [ 1, 1], ,j q EADCAR − + is the cumulative abnormal return for the 3-days surrounding

the earnings announcement to which UE relates. Thus, α2 captures the amount of total earnings

news that is revealed during the earnings announcement period and α3 captures the change in

this relation post-stoppage. If without guidance, investors learn less about the upcoming

earnings surprise prior to the earnings announcement then, presumably, the information content

of the earnings announcement should increase and α3 should be positive.

The results of equation 4a are presented in Table 5, Panel A. For the event sample, the

coefficient on UE is positive and significant, suggesting that the market learns about the upcoming

earnings surprise prior to the earnings announcement. However, after the firm stops providing

27 An alternative would be to measure UE relative to the first forecast issued by an analyst at the beginning of the quarter. However, this data requirement would further reduce our sample.

28

guidance, this “learning” is significantly reduced (α3 is significantly negative, t=-4.75). Moreover,

for the control sample, we do not find a difference for the “post” quarters. The decline in

“learning” is significantly greater for the event sample than for the control sample (t=-5.32).

The results for equation 4b are presented in Table 5, Panel B. Overall, we do not find a

statistically significant change in the information content of the earnings announcement for either

our event or control firms, nor do we find any difference between these two groups of firms.

4.2.2 Volatility and trading volume

The prior results suggest that the information content of earnings announcements

increases once firms stop providing guidance. An alternative way to measure information

content is the volatility of returns and trading volume. We compute Beaver (1968)’s U-statistic

for stock returns and volume around earnings announcements [-1, 0, +1 window] in the pre

versus post guidance windows. Beaver’s (1968) U-statistic is the ratio of the average daily stock

return volatility for the 3 days surrounding the earnings announcement relative to the average

daily stock return volatility in non-earnings announcement periods in the quarter. Also, as per

Beaver (1968), we measure abnormal trading volume (TVOL) as the mean 3-day earnings

announcement trading volume scaled by non-announcement period mean trading volume. The

TVOL measure must be positive, by construction. TVOL measures less than one are indicative

of smaller than non-announcement period volatility while TVOL measures greater than one

suggest that the volatility during the announcement period is larger relative to non-announcement

periods.

Table 6, Panel A reports univariate statistics for these two measures (as well as other

variables examined and discussed in the next sub-sections). Panel B reports the results of a

regression analysis using a similar design as that used for our previous analysis (in section 4.2.1).

29

We again include the Inverse Mills Ratio (IMR) to control for the self-selection bias in our

research setting. For a more parsimonious presentation, we report only the coefficient on POST,

which represents the change in the statistic after the firm stops providing guidance. Contrary to

the results in Table 5, Panel B, we find a significant coefficient on POST for the U-statistic for

our event firms (t-stat=1.79), suggesting an increase in the abnormal announcement period

volatility after a firms stops guidance. This increase is marginally significantly greater than the

increase for the control sample (t=1.52). However, we do not find a similar increase for trading

volume.

We also examine changes in the overall volatility of returns. One could argue that

uncertainty about the firm’s future prospects, a key driver behind the stoppage decision, would

result in higher volatility in stock returns. However, some firms that give up guidance have

counter-argued that stopping guidance drives away transient investors such as hedge funds who

bet against the guided earnings number and thus results in lower stock return volatility (NIRI

2004). To examine this issue, we examine changes in overall return volatility before and after

the stoppage announcement. Specifically, for the pre-event volatility we measure the standard

deviation of returns in the window -90 to -2 days relative to the stoppage announcement and for

the post event volatility we measure the standard deviation of returns in the window +2 to +90

days (STD90).

Results of these tests are also reported in Table 6. We do not find a significant decline in

overall return volatility for the event group (t-stat on POST = -0.77). The control group does

exhibit a significant decline in volatility (t-stat on POST of -3.80) but the difference between the

two groups is not significant. Thus, we do not find any evidence of an unusual change in return

volatility for firms that announce the decision to stop providing earnings guidance.

30

4.2.3 Analyst following, dispersion and accuracy

Finally, we examine changes in three analyst-related factors: 1) analyst following, 2)

dispersion in analyst forecasts, and 3) analyst forecast accuracy. We measure analyst following

(AF) as the number of analysts issuing forecasts in the pre- and post-stoppage quarters. If the

elimination of guidance increases the analyst’s cost of covering a firm, analyst following should

fall for our event firms. Analyst dispersion (DISP) is measured as the standard deviation of

analysts’ forecasts scaled by the absolute value of the corresponding mean forecast. Since

earnings guidance likely increases the amount of public information about a firm, the elimination

of guidance should lead to an increase in forecast dispersion (Barron, Kim, Lim and Stevens,

1998). Moreover, managers often express concerns over increased dispersion as a reason for

maintaining the practice of giving earnings guidance.28 Forecast error (|FE|) is the absolute

value of actual earnings per share less the mean analyst forecast (scaled by the mean analyst

forecast) in the pre- and post-stoppage quarters. If earnings guidance increases the analysts’

information set, we would expect forecast error to increase for our event firms after the

elimination of guidance. On the other hand, if analysts replace the lack of guidance with greater

analysis and therefore, generate additional private information, then the lack of guidance will not

result in a decline in forecast accuracy.

Table 6 presents the results of our analysis. Univariate statistics are presented at the

bottom of Panel A and regression results are presented at the bottom of Panel B. Contrary to

concerns expressed by managers, we do not find a statistically significant decline in analyst

following after the stoppage of guidance for our event firms. However, we do find that the

28 For example, “Some market observers argue that lack of specific EPS or net income guidance will heighten volatility because a wider range of analyst estimates will result” (Harper 2003). Also a NIRI Update article entitled “The Lighthouse: Earnings Guidance” (NIRI, December 2004) cites one reason behind Tyson Foods decision to provide quarterly guidance as, “…giving guidance…leads to a tighter range of analysts’ estimates, which in turn can result in more accurate valuation.”

31

increase in dispersion and forecast error in the POST period experienced by event firms is

significantly greater than that experienced by control firms (t=2.56 and t= 2.08 for dispersion and

accuracy respectively). Thus, it appears analysts are not able to independently generate similar

levels of information in the absence of guidance, resulting in a greater decrease in forecasting

accuracy among analysts of our event firms relative to our control firms.

5. Evidence of changes in other disclosures

As a final test, we examine whether firms change their level of disclosure following their

decision to stop providing guidance. Firms often claim that they will replace specific earnings

guidance with greater qualitative disclosures about the company’s future plans and strategies.

Such a substitution is consistent with calls from industry groups for changes to the practice of

providing quarterly earnings guidance (Krehmeyer and Orsagh 2006).

However, our prior evidence of 1) a negative market reaction to the stoppage

announcement and 2) the increase in forecast dispersion and decrease in forecast accuracy,

suggests that firms have not replaced earnings guidance with adequate qualitative disclosures.

To provide some evidence on this question, we gather the company’s earnings announcement

press release in the quarter prior to the stoppage announcement (q-1) and the quarter following

the stoppage announcement (q+1). We count the number of words in the press release (DISC) as

a measure of the level of disclosure. We then calculate the percent change (%ΔDISC) in the

level of disclosure as (DISCq+1-DISCq-1)/DISCq-1. Of our original sample of 96 firms, we are

able to calculate %ΔDISC for 86 firms.

Panel A of Table 7 provides descriptive statistics on %ΔDISC. Both the mean and

median values are significantly greater than zero, suggesting that firms, on average, have

32

increased their level of disclosure in the quarter following the stoppage announcement.

Approximately 55.8% of firms show an increase in the DISC, while 44.2% show a decrease.

However, two factors potentially impact this comparison. First, the length of a company’s

earnings announcement press release likely changes as the fiscal year progresses. In particular,

firms likely disclose more in later fiscal quarters (Matsumoto, Pronk, and Roelofsen 2006).

Since a large number of firms announced the decision to stop providing earnings guidance in the

first quarter of 2003, it is likely that a disproportionate number of pre-stoppage quarters are

fourth fiscal quarters and post-stoppage quarters are second fiscal quarters. To the extent the

length of press releases in the fourth fiscal quarter is longer than in the second fiscal quarter,

%ΔDISC will be biased downward. In addition, firms also likely disclose more when reporting

poor earnings performance (D’Souza, Ramesh and Shen 2006, Matsumoto et al., 2006). If firm

performance is deteriorating over time, the increase in DISC may be due to performance rather

than the stop guidance decision.

To control for these possible effects, we run the following regression:

0 1 2 3% DISC QTR LOSS MISSβ β β β εΔ = + Δ + Δ + Δ + (5)

ΔQTR is the difference between the q+1 fiscal quarter and the q-1 fiscal quarter. If the

post stoppage quarter is earlier (later) in the fiscal year than the pre stoppage quarter ΔQTR will

be negative (positive). Assuming firms issue longer earnings press releases later in the fiscal

year, the coefficient on ΔQTR should be positive. LOSS is a dummy variable equal to 1 if

earnings before extraordinary items (Quarterly Compustat #8) is less than zero, and zero

otherwise. ΔLOSS is LOSS in quarter q+1 less LOSS in quarter q-1. MISS is a dummy variable

equal to 1 if actual earnings is less than the mean consensus analyst forecasts (both per IBES),

and zero otherwise. Similar to ΔLOSS, ΔMISS is MISS in quarter q+1 less MISS in quarter q-1.

33

If firms issue longer earnings press releases when reporting losses or when they miss analysts’

forecasts, the coefficient on ΔLOSS and ΔMISS should be positive. A significantly positive

coefficient on the intercept would indicate that disclosure levels have increased following the

decision to stop guidance, even after controlling for the effect of fiscal quarters and firm

performance,

Panel A of Table 7 presents descriptive statistics on these variables. As we suspected,

ΔQTR is significantly negative, indicating that post-stoppage quarters tend to be earlier in the

fiscal year than the pre-stoppage quarters. ΔLOSS and ΔMISS are both significantly positive

indicating that firms report losses and miss analysts’ expectations more frequently in the post-

stoppage quarter versus the pre-stoppage quarter. The results of estimating equation 5 are

presented in Panel B of Table 7. We first run the regression including only ΔQTR, as data

requirements to calculate ΔLOSS and ΔMISS reduce our sample size to 53. In both

specifications, the coefficient on ΔQTR is significantly positive, indicating that firms disclose

more in later fiscal quarters. We also find that firms disclose more when reporting losses – the

coefficient on ΔLOSS is significantly positive. Finally, in both specifications we find a

significantly positive intercept, indicating that disclosure levels have indeed increased after

controlling for fiscal quarter effects and firm performance. However, given our findings of

increased analyst forecast dispersion and decreased analyst forecast accuracy, it appears this

increased disclosure is not a direct substitute for forecast guidance.29

29 We repeat this analysis replacing ΔLOSS with the change in earnings between the pre and post quarters, scaled by total assets. The intercept remains significantly positive (the coefficient on the change in earnings is insignificant).

34

6. Conclusions

The well-publicized decision by the Coca-Cola Company to stop providing earnings

guidance to the market, followed by a number of other companies issuing similar statements,

raises questions about the motives behind these companies’ actions as well as the consequences

of their decision. We investigate these issues using a sample of firms that publicly announce

their decision to stop providing quarterly earnings guidance.

Our results suggest that poorly performing firms and firms with low external demand for

guidance are more likely to renounce guidance. However, we do not find evidence that firms

stop guiding when they have a high concentration of investors with longer investment horizons.

Thus, while many firms provide altruistic reasons for no longer providing guidance (e.g., desire

to focus on the long-term), it appears that for many firms the driving factor is poor performance.

We also find that the market reacts negatively to the news that a firm will no longer be

providing earnings guidance. There are two possible explanations for this negative price

reaction: 1) the market interprets the act of stopping as a signal about future performance and

therefore revises downward its expectation about future cash flows or 2) the act of guidance

reduces systematic risk that decreases the cost of equity capital and therefore, the elimination of

guidance increases the firm’s cost of capital. In additional analysis, we find support primarily

for the first explanation – that is, firms with worse future performance suffer a greater market

decline when they announce the decision to stop providing guidance. We find evidence that

firms stopping guidance experience an increase in systematic risk but this change is not

associated with the announcement period return.

Finally, we find that after a firm stops providing guidance, less earnings news is revealed

to the market prior to the earnings announcement date. However, the act of renouncing guidance

35

does not appear to increase return volatility or lower analyst following, contrary to the concerns

often expressed about the costs of foregoing guidance. We also find an increase in analyst

forecast dispersion and analyst forecast error after stoppage of guidance for our event firms

relative to our control firms, despite the fact that, on average, firms disclose more information

after stopping guidance. These findings suggest that explicit guidance is valuable to analysts and

that they are not able to independently generate similar levels of information through private

information acquisition or the interpretation of additional qualitative managerial disclosures.

Overall, our results suggest that guidance is valuable to investors and analysts. It

increases analyst forecast accuracy, reduces dispersion and may even reduce systematic risk.

Managers may want to consider these benefits to guidance when evaluating the decision to stop

providing guidance. Given the documented costs to giving up guidance, one might wonder why

firms made this decision. There are two potential reasons. First, managers may not have

realized the cost associated with discontinuing guidance. Given the flurry of announcements

shortly after Coke’s announcement, there may not have been sufficient time for managers to

realize the consequences of their decision. Second, we do not (and cannot) document the

consequences the firm would have experienced had they continued to provide guidance. For

example, Graham, Harvey and Rajgopal (2005, 42) find that Chief Financial Officers consider

failing to meet or beat a guided earnings target to be worse than not providing earnings guidance

in the first place. It is thus possible that the consequences of continuing with earnings guidance

outweigh the consequences the firm experienced from its decision to discontinue guidance.

36

REFERENCES Anilowski, C., M. Feng, and D. J. Skinner. 2005. Is guidance a macro factor? The nature and

information content of aggregate earnings guidance. Working paper, University of Michigan, University of Pittsburgh, and University of Chicago.

Ashbaugh, H., D. Collins, and R. LaFond. 2005. Corporate governance and the cost of equity

capital. University of Wisconsin and University of Iowa, working paper. Bartov, E., D. Givoly, and C. Hayn. 2002. The rewards to meeting or beating earnings

expectations. Journal of Accounting and Economics 33: 173-204. Barron, O., O. Kim, S. Lim, and D. Stevens. 1998. Using analysts’ forecasts to measure

properties of analysts’ information environment. The Accounting Review 73(3): 421-433.

Beaver, W. 1968. The information content of annual earnings announcements. Journal of

Accounting Research 6 (Supplement): 67-92. Botosan, C. 1997. Disclosure level and the cost of equity capital. The Accounting Review

72(3): 323-349. Botosan, C. and M. Plumlee. 2002. A re-examination of disclosure level and the expected cost

of equity capital. Journal of Accounting Research 40 (1): 21-40. Browning, C. 2006. Statement submitted to the Subcommittee on Capital Markets, Insurance,

and Government Sponsored Enterprises Committee on Financial Services, U.S. House of Representatives at a hearing entitled "Fostering Accuracy and Transparency in Financial Reporting" March 29.

Buffett, W. 1996. An owner’s manual.

http://www.berkshirehathaway.com/ownman.pdf#search=%22warren%20buffet%20earnings%20guidance%22

Chen, S., DeFond, M., and Park, C., 2002. Voluntary disclosure of balance sheet information in

quarterly earnings announcements. Journal of Accounting and Economics 33: 229-251. Cheng, M., K.R. Subramanyam and Y.Zhang. 2005. Earnings guidance and managerial myopia.

Working paper, University of Southern California. Cremers, K. J. Martijn and V. B. Nair. 2004. Governance mechanisms and equity prices.

Working paper, Yale University and New York University. Cook, T. D. and D. T. Campbell. 1979. Quasi-Experimentation: Design and Analysis Issues for

Field Studies. Boston, MA: Houghton Mifflin Company.

37

D’Souza, J., K. Ramesh, and M. Shen. 2006. Disclosure of financial statement information in earnings announcements. Working paper, Cornell University, Michigan State University and George Mason Universtiy.

Del Guercio, D. and J. Hawkins. 1999. The motivation and impact of pension fund activism.

Journal of Financial Economics 52: 293-340. Dye, R. 2001. An evaluation of ‘essays on disclosure’ and the disclosure literature in

accounting. Journal of Accounting & Economics 32, nos. 1-3: 181-236. Fama, E., K. French, 1993. Common risk factors in the returns on stocks and bonds. Journal of Financial Economics 33: 3-56.

Francis, J., R. LaFond, P. Olsson, and K. Schipper. 2004. Cost of equity and earnings attributes.

The Accounting Review 79(4): 967-1010. Francis, J., D. Philbrick, and K. Schipper. 1994. Shareholder litigation and corporate disclosure.

Journal of Accounting Research 32(2): 137-164. Fuller, J., and M. Jensen. 2002. Just say no to Wall Street. Journal of Applied Corporate Finance

14 (4): 41-46. Graham, J., C. Harvey and S.Rajgopal. 2005. The economic implications of financial reporting.

Journal of Accounting and Economics 40:3-73. Greene, W. H. 1997. Econometric Analysis (third edition). Prentice-Hall, Inc., Upper Saddle

River, NJ. Grullon, G., R. Michaely, and B. Swaminathan. 2002. Are dividend changes a sign of firm

maturity? The Journal of Business 75(3): 387-424. Grullon, G. and R. Michaely. 2004. The information content of share repurchase programs. The

Journal of Finance 59(2): 651-680. Harper, H. 2003. Don’t give up guidance. Strategic Investor Relations (Spring): 41-47. Healy, Paul M., and Krishna G. Palepu. 2001. Information asymmetry, corporate disclosure, and

the capital markets: A review of the empirical disclosure literature. Journal of Accounting & Economics 31, nos. 1-3: 405-440.

Healy, P., A. Hutton, and K. Palepu. 1999. Stock performance and intermediation changes

surrounding sustained increases in disclosure. Contemporary Accounting Research 16(3): 485-520.

38

Houston, J., B. Lev, and J. Tucker. 2006. To guide or not to guide? Causes and consequences of stopping and subsequently resuming earnings guidance. University of Florida and New York University, working paper.

Hutton, A. 2005. Determinants of managerial earnings guidance prior to Regulation Fair

Disclosure and bias in analysts’ earnings forecasts. Contemporary Accounting Research 22(4): 867-914.

Jensen, M., K. Murphy and E. Wruck. 2004. Remuneration: Where we’ve been, how we got to

here, what are the problems, and how to fix them. Working paper, Harvard Business School.

Krehmeyer, D. and M. Orsagh. 2006. Breaking the short-term cycle: Discussion and

recommendations on how corporate leaders, asset managers, investors, and analysts can refocus on long-term value. Proceedings of the CFA Center for Financial Market Integrity and Business Roundtable Institute for Corporate Ethics symposium series on short-termism.

Lang, M. and R. Lundholm. 1993. Cross-sectional determinants of analysts’ ratings of corporate

disclosures. Journal of Accounting Research 31 (2): 246-270. Matsumoto, D., M. Pronk, and E. Roelofsen. 2006. Managerial disclosure vs. analyst inquiry: an

empirical investigation of the presentation and discussion portions of earnings-related conference calls. Working Paper, University of Washington, Tilburg University, and RSM/Erasmus University.

Miller, G. S. 2002. Earnings performance and discretionary disclosure. Journal of Accounting

Research 40(1): 173-204. Morgan, G. 2003. Earnings guidance: The good, the bad and the ugly. Wall Street Lawyer. Also

available at www.glasserlegalworks.com. National Investor Relations Institute (NIRI). 2003. NIRI survey results on earnings guidance

practices. April 8. National Investor Relations Institute (NIRI). 2004. Best Practices: Earnings guidance, or no

earnings guidance? October 5. Rogers, J. and P. Stocken. 2005. Credibility of management forecasts. Accounting Review 80(4):

1233-1260. Securities and Exchange Commission (SEC). 2000. Selective Disclosure and Insider Trading:

Proposed rule S7-31-99. Available at http://www.sec.gov/rules/proposed/34-42259.htm

39

Skinner, D. and R. Sloan. 2002. Earnings surprises, growth expectations, and stock returns or don’t let an earnings torpedo sink your portfolio. Review of Accounting Studies 7(2-3): 289.

Tasker, S. 1998. Bridging the information gap: Quarterly conference calls as a medium for

voluntary disclosure. Review of Accounting Studies 3 (1-2): 137-167. Thompson, L., 2003a. NIRI survey results on earnings guidance practices. Executive Alert,

National Investor Relations Institute, April 8, 2003. Thompson, L., 2003b. NIRI issues survey results on earnings guidance practices. Executive

Alert, National Investor Relations Institute, December 10, 2003. Thompson, L., 2005. NIRI issues survey results on earnings guidance practices. Executive Alert,

National Investor Relations Institute, March 30, 2005.

Figure 1 Distribution of Announcement Dates for Stop Guidance Sample

0

2

4

6

8

10

12

14

Jan-0

1Mar-

01May

-01Ju

l-01

Sep-01

Nov-01

Jan-0

2Mar-

02May

-02Ju

l-02

Sep-02

Nov-02

Jan-0

3Mar-

03May

-03Ju

l-03

Sep-03

Nov-03

Jan-0

4Mar-

04May

-04Ju

l-04

Sep-04

Nov-04

Jan-0

5Mar-

05May

-05Ju

l-05

Sep-05

Nov-05

Jan-0

6

Date of Adoption

No

of fi

rms

COKE

41

Figure 2 Time Line for Defining Variables used in Future ERC Tests

(Panel A of Table 5)

Earnings Announcement Quarter Q-2

Earnings AnnouncementQuarter Q-1

Earnings AnnouncementQuarter Q

Earnings Announcement Quarter Q+1

Event quarter Q Post-event quarter Pre-event quarter

Day +2 Day -2 Day +2 Day -2

CAR regressed on UE of pre-event quarter Q-1

CAR regressed on UE of post-event quarter Q+1

42

Table 1 Sample Attrition for Firms that Stop Quarterly Guidance

Sample Selection Criteria Number of observations left Test of

Determinants Test of Market

Reaction Test of

Consequences Initial Sample 96 firms 96 firms 96 firms Less: 6 firms not on CRSP (5 of which are foreign firms) 90 firms 90 firms 90 firms Less: 4 firms deleted from CRSP before announcement date 86 firms 86 firms 86 firms Less: 2 firms without CRSP data on BH_1, BH_2, STDret 84 firms 84 firms 84 firms Less: 4 firms without Compustat data on PMBH, PLOSS, LNMV_1, and MB

80 firms 80 firms 80 firms

Less: 5 firm without Compustat and CRSP data on PRED, VALREL, and litigation probability LIT

75 firms 75 firms 75 firms

Less: 2 firms with no data on consensus analyst forecast before stop announcement dates on IBES

73 firms

Less: firms with no CRSP data for estimation of change in beta and no IBES forecast data for calculating analyst forecast revision

53-63 firms

Less: firms without requisite IBES, Compustat and/or CRSP data (due to further restriction on having quarter Q+1 data)

53-66 firms

43

Table 2 Determinants of the Decision to Stop Quarterly Guidance

Panel A: Univariate Statistics Mean Median Test of Difference Event

(n=75) Control (n=901)

Event (n=75)

Control (n=901)

t-statistics z-statistics

BH_1 -0.10 0.17 -0.13 0.04 -3.77*** -5.23*** BH_2 0.32 0.26 0.06 0.15 0.56 -1.06 PLOSS 0.30 0.23 0.13 0.13 1.99** 2.03** PMBAF 0.80 0.84 0.88 0.88 -1.99** -1.91** VALREL -0.20 -0.16 -0.06 -0.03 -0.56 -3.33*** STDret 0.03 0.03 0.03 0.03 0.84 -0.02 PRED -0.05 -0.48 -0.02 -0.02 -1.83** -0.42 AF 7.48 9.18 5 8 -2.05** -2.58*** PINST 57.96 68.46 64.10 72.88 -3.39*** -3.50*** LIT 0.02 0.05 0.002 0.002 -1.88** 1.84** Db 0.91 0.88 0 1 0.55 0.55 BOa 37.58 39.21 32.19 37.03 -0.58 -0.58 Dp 0.50 0.57 1 1 -1.06 -1.07 POb 2.82 2.94 2.83 2.98 -0.68 -0.21 LNMV 6.86 6.99 6.53 6.84 -0.55 -1.11 MB 2.69 2.68 1.76 2.31 0.01 -2.56*** LNCT 2.13 2.73 2.48 2.77 -4.50*** -4.04***

44

Table 2 Determinants of the Decision to Stop Quarterly Guidance

(continued)

Panel B: Correlation Matrix for Test Variables. Pearson Correlation Coefficients above the Diagonal and Spearman Rank Correlation Coefficients below the Diagonal

BH_1 BH_2 PLOSS PMBAF VALREL STDret PRED AF PINST LIT BO PO LNMV MB LNCT BH_1 0.01 -0.01 0.06** -0.09*** 0.02 0.01 -0.15*** -0.19*** -0.10*** 0.04 -0.05 -0.05* 0.06** -0.02 BH_2 -0.09*** -0.14*** 0.17*** -0.16*** 0.05 0.00 -0.12*** -0.05 0.05 0.02 0.01 -0.10*** 0.02 -0.03 PLOSS -0.13*** -0.28*** -0.14*** -0.11*** 0.59*** 0.02 -0.12*** -0.16*** 0.00 -0.04 -0.17*** -0.32*** -0.02 -0.17*** PMBAF 0.07** 0.18*** -0.09*** 0.02 -0.04 0.06* 0.05* 0.03 0.03 -0.07** -0.06** 0.06* 0.07** 0.04 VALREL 0.01 0.04 -0.13*** 0.00 -0.20*** 0.00 0.07** 0.13*** -0.10*** -0.01 0.07** 0.11*** 0.01 0.13*** STDret -0.15*** -0.08*** 0.52*** -0.02 -0.26*** 0.02 -0.11*** -0.20*** 0.20*** -0.05* -0.21*** -0.46*** -0.02 -0.22*** PRED 0.02 0.09*** -0.45*** 0.01 0.10*** -0.30*** -0.04 -0.07** 0.01 0.02 -0.00 -0.06** 0.00 -0.09*** AF -0.11*** -0.09*** -0.13*** 0.06* 0.07** -0.15*** 0.19*** 0.26*** 0.11*** 0.04 0.13*** 0.67*** 0.04 0.26*** PINST -0.05 0.07** -0.13*** 0.07* 0.11*** -0.18*** 0.08** 0.31*** 0.05* 0.03 0.06** 0.28*** -0.04 0.28*** LIT -0.44*** 0.07** 0.06** 0.04 -0.12*** 0.32*** 0.05 0.54*** 0.24*** 0.02 -0.06** 0.03 0.00 -0.04 BO 0.04 -0.00 -0.04 -0.05 0.08** -0.05* -0.00 0.02 0.05 -0.01 0.15*** 0.03 0.04 0.03 PO -0.01 0.09*** -0.19*** -0.08** 0.05 -0.23*** -0.07** 0.14*** 0.08** 0.03 0.16*** 0.21*** -0.04 0.14*** LNMV 0.09*** -0.03 -0.33*** 0.04 0.11*** -0.50*** 0.30*** 0.71*** 0.28*** 0.36*** 0.06* 0.25*** 0.07** 0.28*** MB 0.31*** 0.07** -0.27*** 0.13*** -0.00 -0.22*** 0.24*** 0.33*** 0.08** 0.14*** 0.01 -0.02 0.48*** -0.04 LNCT 0.00 0.03 -0.12*** 0.05 0.13*** -0.22*** 0.03 0.29*** 0.24*** 0.09*** 0.05 0.16*** 0.32*** 0.06*

Table 2 Determinants of the Decision to Stop Guidance Quarterly Guidance

(continued) Panel C: Logistic Regressionsc Dependent variable = 1 if the firm announces stopping guidance (n=75) 0 for control firms (n=901). Model:

0 1 2 3 4 5 6 7 8

9 10 11 12 13 14 15 16 17 18

_1 _ 2 ret

b b p p

STOP BH BH PLOSS PMEET VALREL STD PRED AFHIAF AF PINST LIT D D BO D D PO LNMV MB LNCT

β β β β β β β β ββ β β β β β β β β β ε

= + + + + + + + +

+ × + + + + × + + × + + + +

Predicted

sign Raw Values for RHS

variables Ranked Values for RHS

variables Independent variables

Coefficients p-values Coefficients p-values BH_1 − -1.68 <0.01 -2.11 <0.01 BH_2 − 0.05 0.24 -0.55 0.14 PLOSS + 0.49 0.19 1.01 0.06 PMBAF − -1.07 0.07 -0.52 0.14 VALREL − -0.03 0.45 -1.45 <0.01 STDret − -1.98 0.45 -2.67 <0.01 PRED + 0.48 0.34 -0.09 0.15 AF − -0.16 <0.01 -2.62 <0.01 HI×AF + 0.08 0.02 1.54 0.07 PINST − -1.26 0.03 -1.31 <0.01 LIT + -0.84 0.68 1.85 <0.01 Db ? 1.19 0.04 0.98 0.08 Db×BO + -0.58 0.64 -0.26 0.38 Dp ? 0.02 0.98 -0.18 0.76 Dp×PO + -0.17 0.68 -0.60 0.44 LNMV ? 0.38 <0.01 0.71 0.44 MB ? -0.00 0.98 -0.45 0.42 LNCT - -1.09 <0.01 -1.51 <0.01 Pseduo-R2

19.13

17.86

46

Table 2 Determinants of the Decision to Stop Quarterly Guidance

(continued) Notes to Table 2: *, **, ***: significant at 10%, 5%, 1% (two-sided p-values) for Panel B correlation matrix. a For the BO variable we report the mean and median values only for observations with requisite data from Compact Disclosure. The sample sizes are 68 for the event observations and 798 for the control observations. b For the PO variable we report the mean and median values only for observations with requisite data from Cremers and Nair (2004). The sample sizes are 38 for the event observations and 514 for the control observations. cP-values for Panels A and C are one-sided p-values for variables with directional predictions. We boldface coefficients with p-values less than 10%. We report (1-p) values for coefficients that assume a sign opposite to the one predicted. Definition of variables: BH_1 = market-adjusted buy-and-hold returns for the 12 months beginning from month -12 ending month -1, with month 0 being the announcement month. BH_2 = market-adjusted buy-and-hold returns for the 12 months beginning from month -24 ending month -13, with month 0 being the announcement month. PLOSS = percentage of losses in the 8 quarters proceeding the announcement quarter. Loss is coded as 1 if net

income (Compustat quarterly #19) is negative. We require at least 4 quarters of non-missing net income on Compustat to compute this variable.

PMBAF = percentage of meeting or beating earnings expectations in the 8 quarters preceding the announcement quarter. Expected earnings is measured as consensus analyst forecasts before earnings announcements from IBES. We require at least 4 quarters of non-missing net income on Compustat to compute this variable.

VALREL = value relevance proxy, measured as negative one times the squared residual from the following regression: 0 1 2 3 4*RET NIBE LOSS NIBE LOSS NIBEβ β β β β ε= + + + + Δ + . The regression is estimated by three, two, or one-digit SIC code conditional on having at least 10 firms in each group. RET = market adjusted return over the fiscal year preceding guidance stoppage; NIBE = net income before extraordinary items (Compustat annual data18) scaled by the beginning market value of equity (Compustat annual #25*Compustat annual #199); LOSS =1 if NIBE is negative, zero otherwise; ΔNIBE = the change in net income before extra-ordinary items scaled by beginning market value of equity. Since we multiply by -1 larger values of VALREL indicates greater value relevance. NIBE is computed over the fiscal year preceding guidance stoppage.

STDret = standard deviation of daily raw returns during 252 days prior to the stop-guidance announcement date for event observations and the management forecast date for control observations.

PRED = predictability of earnings, measured as negative one times the square root of the error variance from firm j’s AR(1) model: ΔNIq = α+βΔNIq-1+ε, where ΔNIq is seasonally adjusted net income (Compustat quarterly #69 scaled by beginning market value of equity). We require at least 8 observations per firm in estimating PRED. Since we multiply by -1 larger values of PRED indicates greater predictability. NI comes from fiscal periods preceding quarter of guidance stoppage.

AF = number of analysts following the firm, measured as the number of analysts covering the firm at the beginning of the event quarter per IBES. If a firm is not on IBES, we code AF as zero. The window over the 252 days preceding to the stop-guidance announcement (management forecast) date for event (control) firms.

47

Table 2 Determinants of the Decision to Stop Quarterly Guidance

(continued)

HIAF = an indicator variable for the 4th quartile of AF. PINST = percentage institutional ownership measured prior to the announcement date as reported on Compact

Disclosure. LIT = proxy for litigation risk, calculated using the Rogers and Stocken (2005) litigation probability model

coefficients estimates. Larger values indicate greater probability of litigation. Specifically, probability (litigation=1) = G(-5.378+0.141*SIZE + 0.284*TURNOVER + 0.012*BETA -0.237*RETURNS -1.340*STDRET -0.011*SKEWNESS -3.161*MINRET -0.025*BIOTECH + 0.378* HARDWARE + 0.075*ELECTRONICS -0.034*RETAIL + 0.211*SOFTWARE, where G is the standard normal cumulative distribution function. The estimation of all the independent variables (except for the industry dummies) use the window over 252 days preceding the stop guidance announcement (management forecast) date for event (control firms). Note that SIZE = the natural log of the average market value of equity; TURNOVER = average daily share volume divided the average shares outstanding; BETA = the slope coefficient from regressing firm’s daily returns on CRSP Value-Weighted Index; RETURNS = the cumulative daily raw stock returns; SKEWNESS = the skewness of daily returns; MINRET = the minimum of daily returns. The high risk industry dummies represent Biotech (SIC 2833 to 2836), Computer Hardware (SIC 3570 to 3577), Electronics (SIC 3600to 3674), Retailing (SIC 2500 to 5961) and Computer Software (SIC 7371 to 7379). For presentation purposes we multiply the estimated probability by 102.

Db = indicator variable coded as 1 if the firm has 5% blockholder ownership data from Compact Disclosure. BO = percentage 5% blockholder ownership of the firm prior to stop-guidance announcement from Compact

Disclosure. Db×BO = interaction variable between Db and BO. Dp = indicator variable coded as 1 if the firm has pension ownership data. PO = percentage pension ownership of the firm prior to stop-guidance announcement. Data source: Cremers and

Nair (2004). Dp×PO = interaction variable between Dp and PO. LNMV = natural log of the beginning of event quarter market value of equity (Compustat quarterly #14*#61). MB= natural log of the beginning of event quarter market to book ratio. LNCT = natural log of (1+ CT), where CT is the number of management quarterly EPS forecasts up till quarter t−1.

48

Table 3 Market Reaction to Announcement of Stopping Quarterly Guidance (N=73)

Panel A: Univariate Statistics MEAN MEDIAN %

negative Dea= 1 (n=54)

Dea= 0 (n=19)

t-stat for mean

difference

Dea= 1 (n=54)

Dea= 0 (n=19)

z-stat for median

differenceUE 30.1% -0.013 N/A N/A 0.000 N/A N/A CAR-1,1 65.8% -0.027 -0.048 -0.87 -0.020 -0.028 -0.87 RET-1,1 65.8% -0.028 -0.052 -0.92 -0.017 -0.024 -0.91 Panel B: OLS Regression:

1,1 1 2( ) ( )ea eaCAR D D UEα β β ε−

= + + × +

Independent variables Coefficients t-statistic Intercept -0.048 -2.31 Dea 0.024 0.97 Dea×UE 0.187 1.21 Adjusted R2

3.16%

Notes to Table 3: Definition of variables: Dea=indicator variable coded as 1 if an earnings announcement falls in [-1,+1] window centered on the stop-guidance announcement date, 0 otherwise. UE = unexpected earnings, defined as actual EPS minus the most recent preceding consensus analyst forecast on

IBES, scaled by price two days before the earnings announcement date. CAR-1,1 = three-day cumulative abnormal returns centered on the stop-guidance announcement date. RET-1,1 = three-day cumulative raw returns centered on the stop-guidance announcement date.

49

Table 4 Further Analysis of Market Reaction – Cost of Capital vs. Future Performance

Panel A: Test of shift in CAPM beta Model [ ] [ ]it ft post i i mt ft post i mt ft itR R POST R R POST R Rα α β β ε− = + + − + × − +

postβ [ ]post i iCBETA POSTβ β= ÷ Mean Median Mean Median Event 0.552 0.455 2.582 0.408 (p<0.01) (p<0.001) (p<0.01) (p<0.001) Control 0.171 0.247 -0.512 0.157 (p<0.001) (p< 0.001) (p<0.001) (p<0.001) Test of Difference

t = 2.40 z = 2.11 t = 1.85 z = 2.41

Panel B: Determinants of market reaction to announcement of stopping guidance

1,1 1 2 3 4 5 6

7 8 9 10 11

( ) ( ) _1ea eaCAR D D UE FEARN RCBETA BH PMBAF

AF PINST LOPRED LTFOCUS SIZE

α β β β β β β

β β β β β ε−

= + + × + Δ + + +

+ + + + + +

Independent variables

Model (1)

Model (2)

Model (3)

Model (4)

Model (5)

Intercept -0.048 (-2.14)

-0.057 (-1.81)

-0.046 (-1.60)

-0.154 (-2.19)

-0.109 (-1.17)

Dea 0.064 (2.43)

0.026 (0.96)

0.065 (2.38)

0.069 (2.53)

0.057 (2.12)

Dea ×UE 1.975 (1.53)

0.178 (1.09)

1.978 (1.52)

1.905 (1.46)

1.802 (1.43)

ΔFEARN 2.719 (3.83)

2.719 (3.79)

2.579 (3.29)

2.371 (2.72)

RCBETA 0.002 (0.39)

-0.000 (0.08)

0.000 (0.19)

0.000 (0.01)

BH_1 -0.016 (-0.72)

-0.019 (-0.92)

PMBAF 0.130 (2.20)

0.118 (2.08)

AF -0.000 (-0.95)

-0.000 (-0.41)

PINST 0.013 (0.28)

0.005 (0.10)

LOPRED -0.023 (-0.92)

LTFOCUS 0.057 (2.27)

SIZE -0.005 (-0.52)

N 53 63 53 53 53 Adjusted R2 32.85% 3.98% 31.46% 33.56% 39.63%

50

Table 4 Further Analysis of Market Reaction – Cost of Capital vs. Future Performance

(continued) Notes to Table 4: Definition of variables: Panel A: Rit = monthly stock returns for firm i. Firms with less than 6 months of return data after the event date are dropped. Rft = monthly risk-free rates Rmt = monthly return on the NYSE-AMEX-NASDAQ value-weighted market portfolio POST = indicator variable coded as 1 for the months after the stop-guidance announcement date, and 0 otherwise. Panel B: Dea=indicator variable coded as 1 if an earnings announcement falls in [-1,+1] window centered on the stop-guidance announcement date, 0 otherwise. UE = unexpected earnings, defined as actual EPS minus the most recent preceding consensus analyst forecast on

IBES, scaled by price two days before the earnings announcement date. CAR-1,1 = three-day cumulative abnormal returns centered on the stop-guidance announcement date. RET-1,1 = three-day cumulative raw returns centered on the stop-guidance announcement date. ΔFEARN = measure of expected future earnings, calculated as analyst forecast revision for quarter Q+1 (analyst

EPS forecast for quarter Q+1 made in quarter Q+1, minus analyst EPS forecast for quarter Q+1 made in quarter Q-1, scaled by price at the end of the quarter before stop-guidance announcement (Q-1).

CBETA= % change in β estimated as postβ / β from firm-specific CAPM model: [ ] [ ]jt ft post mt ft post mt ft jtR R POST R R POST R Rα α β β ε− = + + − + × − + , where Rjt is firm’s daily stock

returns, and Rit, Rft, Rmt, and POST are defined as above in notes to Panel A. RCBETA= decile values of CBETA BH_1 = market-adjusted buy-and-hold returns for the 12 months beginning from month -12 ending month -1, with month 0 being the announcement month. PMBAF = percentage of meeting or beating earnings expectations in the 8 quarters preceding the announcement

quarter. Expected earnings is measured as consensus analyst forecasts before earnings announcements from IBES. We require at least 4 quarters of non-missing net income on Compustat to compute this variable.

AF = number of analysts following the firm, measured as the number of analysts covering the firm at the beginning of the event quarter per IBES. If a firm is not on IBES, we code AF as zero. The window over the 252 days preceding to the stop-guidance announcement (management forecast) date for event (control) firms.

PINST = percentage institutional ownership measured prior to the announcement date as reported on Compact Disclosure.

LOPRED = one if management’s reason for discontinuing guidance is due to difficulty predicting future earnings/increased uncertainty (as reported in company press release) and zero otherwise.

LTFOCUS = one if management’s reason for discontinuing guidance is to focus on the long-term prospects of the company (as reported in company press release) and zero otherwise.

LNMV = natural log of the beginning of quarter market value of equity (Compustat quarterly #14*#61).

Table 5 Consequence of Stopping Quarterly Guidance: Future ERCs and ERCs

Panel A: Future ERC Test Model [ 2, 2]

, 0 1 2 , 3 , 4 , 1 ,( )j q j j q j j q j q j qCAR POST UE POST UE IMRα α α α α ε+ −−= + + + × + +

Event Sample

Control Sample

Difference

(Stacked Sample) Pred.

sign Coeffs. T-stats Pred.

Sign Coeffs. T-stats. Pred.

Sign Coeffs. T-stats.

Intercept 0.08 (1.31) 0.10* (1.52) -0.03 (-1.03) POST ? 0.03 (0.58) ? 0.03** (2.12) ? -0.00 (-0.14) UE + 5.31*** (4.58) + 0.13* (1.47) ? 4.33*** (5.65) POST*UE − -4.91*** (-4.75) ? 0.16 (0.33) − -4.41*** (-5.32) IMR ? 0.18*** (2.50) ? 0.04 (0.56) ? 0.11*** (2.45)

52

Table 5: Consequence of Stopping Guidance: Future ERCs and ERCs (continued) Panel B: ERC Test Model [ 1, 1]

, , 0 1 2 , 3 , 4 , 1 ,( )j q EAD j j q j j q j q j qCAR POST UE POST UE IMRα α α α α ε− +−= + + + × + +

Event Sample

Control Sample

Difference (Stacked Sample)

Pred. sign

Coeffs. T-stats Pred. Sign

Coeffs. T-stats. Pred. Sign

Coeffs. T-stats.

Intercept -0.02 (-0.43) 0.01*** (2.72) -0.04* (-1.54) POST ? 0.01 (0.62) ? -0.00 (-1.28) ? 0.02 (1.05) UE + 0.02 (0.80) + 0.14*** (3.78) ? 0.04 (0.19) POST*UE + -0.19 (-0.71) ? -0.06 (-0.83) + -0.11 (-0.48) IMR ? 0.02 (0.70) ? 0.04* (1.83) ? 0.02** (1.82)

Notes to Table 5: For the time-line used in this test, please refer to Figure 2. The future ERC tests in Panel A regress returns on upcoming unexpected earnings. The ERC tests in Panel B regress 3-day earnings announcement CAR on unexpected earnings for the quarter. *, **, ***: significant at 10%, 5%, 1%, one-sided. Definition of variables:

[ 2, 2],j qCAR + − = cumulative abnormal returns from two days after the earnings announcement date of the previous quarter to two days before the earnings

announcement date of the upcoming quarter. We obtain earnings announcement dates from Compustat quarterly file (RDQE), and measure this cumulative abnormal return for the pre-event quarter and the post-event quarter (see Figure 1 for more details).

[ 1, 1], ,j q EADCAR − + = 3-day cumulative abnormal returns centered on earnings announcement date.

POSTj = indicator variable coded as 1 for the post-event quarter: for event firms it is the quarter after the stop guidance quarter and for control firms it is the quarter after the matched management forecast quarter.

UEj,q = unexpected earnings measured as seasonally differenced EPS (Compustat quarterly #19), scaled by stock price at the end of quarter Q-2. We measure this unexpected earnings variable for the pre-event and post-event quarters.

,j j qPOST UE× = interaction between POSTj and UEj,q. IMR j,q= Inverse Mills Ratio obtained from the determinant test in Equation (1).

53

Table 6 Consequence of Stopping Guidance: Beaver’s U, Abnormal Trading Volume, and Return Volatility

For the Pre-Event Quarter (Q-1) and Post-Event Quarter (Q+1) Panel A Mean and median values for Beaver’s U statistics, abnormal trading volume, and return volatility

Event Control MEAN MEDIAN MEAN MEDIAN Beaver’s U-statistic Event N=65 Q-1 3.88 1.88 4.52 1.81 [U] Control N = 726 Q+1 5.18 2.06 5.15 1.97 Abnormal Trading Volume Event N=65 Q-1 2.07 1.45 1.82 1.49 [TVOL] Control N = 726 Q+1 1.96 1.63 1.99 1.58 90-day return volatility Event N=66 Q-1 0.03 0.03 0.03 0.03 [STD90] Control N = 838 Q+1 0.03 0.03 0.03 0.03 Analyst Following Event N = 55 Q-1 7.93 5 9.59 8 [AF] Control N = 793 Q+1 7.38 5.5 9.62 8 Forecast Dispersion Event N =53 Q-1 0.24 0.08 0.12 0.04 [DISP] Control N =764 Q+1 0.34 0.10 0.11 0.05 | Forecast Error | Event N =55 Q-1 0.43 0.09 0.27 0.07 [|FE|] Control N =783 Q+1 0.73 0.14 0.23 0.13

54

Table 6 Consequence of Stopping Guidance: Beaver’s U, Abnormal Trading Volume, and Return Volatility

For the Pre-Event Quarter (Q-1) and Post-Event Quarter (Q+1) (continued)

Panel B Regression Results -- Model:

, , , , , , 0 1 2 , 1 ,[ , 90 , , ,| | ]j q j q j q j q j q j q j j q j qU TVOL STD AF DISP FE POST IMRα α α ε−= + + + Event Sample Control Sample Difference between

Event and Control Dependent

variable Coeff. T-stats. Coeff. T-stats. Coeff. T-stats

U POST 5.85** (1.79) 1.55** (1.97) 4.31* (1.52)

TVOL POST 0.36 (1.13) 0.05 (0.55) 0.31 (0.91)

STD90 POST -0.00 (-0.77) -0.00*** (-3.80) 0.00 (0.07)

AF POST 0.00 (0.00) -0.00 (-0.00) 0.00 (0.00)

DISP POST 0.21** (1.95) -0.02 (-1.33) 0.22*** (2.56)

|FE| POST 0.15* (1.71) -0.06 (-1.17) 0.23** (2.08)

55

Table 6 Consequence of Stopping Guidance: Beaver’s U, Abnormal Trading Volume, and Return Volatility

For the Pre-Event Quarter (Q-1) and Post-Event Quarter (Q+1) (continued)

Notes to Table 6: For parsimony we report the coefficients on POST only and do not present the intercepts and the coefficients on IMR (the Inverse Mills Ratio). *,**,***: significant at 10%, 5%, and 1%, one-sided. Definition of variables:

U = Beaver’s U statistics, 2,

2 ( )q t

q

eeσ

. The numerator of Beaver’s U is the mean squared error of the 3-day market-adjusted returns centered on earnings

announcement date, and the denominator is the variance of daily market-adjusted returns of the non-announcement period, where the non-announcement period is defined as the window from two days prior to last quarter’s earnings announcement to two days before current quarter’s earnings announcement date. For event firms the event quarter is the stop-guidance announcement quarter, and for control firms the ‘event’ quarter is the matched management forecast quarter.

TVOL =abnormal trading volume, calculated as the mean 3-day earnings announcement window trading volume scaled by non-announcement period mean trading volume, where the non-announcement period is defined as the window from two days prior to last quarter’s earnings announcement to two days before current quarter’s earnings announcement. For event firms the event quarter is the stop-guidance announcement quarter, and for control firms the ‘event’ quarter is the matched management forecast quarter.

STD90 = standard deviation of daily stock returns for the [-90, -2] window pre-event date and for the [+2, +90] window post-event date, where day 0 is the stop guidance announcement date for event firms and the matched management forecast date for control firms.

AF = analyst following, measured as the number of analysts giving forecasts for the upcoming quarter, in the quarter before and after the event quarter. For event firms the event quarter is the stop-guidance announcement quarter, and for control firms the ‘event’ quarter is the matched management forecast quarter.

DISP = dispersion of analyst forecasts in the quarter before and after the event quarter, measured as the standard deviation of analyst forecasts for upcoming quarter divided by the absolute value of the corresponding mean forecast. For event firms the event quarter is the stop-guidance announcement quarter, and for control firms the ‘event’ quarter is the matched management forecast quarter.

|FE|=accuracy of analyst forecasts in the quarter before and after the event quarter, measured as the absolute value of [(actual EPS – mean analyst forecast/mean analyst forecast]. Because larger |FE| indicates greater deviation of forecast from actual, a bigger number means less accurate. For event firms the event quarter is the stop-guidance announcement quarter, and for control firms the ‘event’ quarter is the matched management forecast quarter.

IMR j,q-1= Inverse Mills Ratio obtained from the determinant test in Equation (1).

56

Table 7 Changes in Length of Disclosures in Earnings Announcement Press Releases

Between the Pre-Event Quarter (Q-1) and Post-Event Quarter (Q+1)

Panel A: Univariate Statistics Std N Mean DeviationMedian %ΔDISC 86 0.0853 0.3399 0.0193 p-valuea 0.0112 0.0700 ΔQTR 86 -0.3605 1.8529 -2.0000 p-value 0.0748 0.0333 ΔLOSS 63 0.1429 0.3958 0.0000 p-value 0.0057 0.0117 ΔMISS 62 0.1935 0.5680 0.0000 p-value 0.0094 0.0071

Panel B: Multivariate Regression 0 1 2 3% DISC QTR LOSS MISSβ β β β εΔ = + Δ + Δ + Δ +

Pred. Model 1 (n=86) Model 2 (n=53) Sign Coefficent t-stat Coefficent t-stat Intercept + 0.1081 3.06 0.0864 2.30 ΔQTR + 0.0632 3.36 0.0760 4.17 ΔLOSS + 0.1741 2.08 ΔMISS + -0.0708 -1.12 R2 =0.1083 R2 =0.2703

Notes to Table 7: a P-values are from t-tests (for mean) and signed ranked tests (for median) testing whether the variable is greater than zero. Variable Definitions: %ΔDISC = The total words disclosed in the company’s earnings press release in quarter q+1 (DISCq+1) less

the total words disclosed in the company’s press release in quarter q-1 (DISCq-1), divided by DISCq-1. ΔQTR = The fiscal quarter of quarter q+1 less the fiscal quarter of quarter q-1 ΔLOSS = LOSS in quarter q+1 less LOSS in quarter q-1, where LOSS is a dummy variable equal to 1 if the

company’s earnings before extraordinary items (Quarterly Compustat #8) is less than zero and zero otherwise.

ΔMISS = MISS in quarter q+1 less MISS in quarter q-1, where MISS is a dummy variable equal to 1 if the company’s actual earnings is less than the mean analyst forecast per IBES.